Use spark sql query elasticsearch sum group by value isn't accurate

[zrk]

When query elasticsearch， I use Spark sql to query elasticsearch，My Spark SQL is： “select log_date,domain,firm_name,sum(traffic) as traffic,concat_ws(':',collect_set(file_name)) as file_name from traff group by log_date,domain,firm_name”，the program can run successfully，but the sum value is not accurate，How can I get the accurate value。